Future FW versions for NVIDIA NICs and DPUs will allow SubFunctions (SFs) to be group into local E-Switch instances. One of such SFs would serve as a transfer proxy for that E-Switch while others will be representors. In this topology, probing in mlx5 networking driver must support working with associated SF devices.
This patch addresses that by adding necessary check for multiport IB device on auxiliary device probing. If multiport device is detected, it'll be treated as E-Switch instance. Probing will be allowed if and only if probed device belongs to set of network devices associated with given IB device. Signed-off-by: Dariusz Sosnowski <[email protected]> Acked-by: Viacheslav Ovsiienko <[email protected]> --- drivers/net/mlx5/linux/mlx5_ethdev_os.c | 11 +- drivers/net/mlx5/linux/mlx5_os.c | 162 +++++++++++++++++++++--- drivers/net/mlx5/linux/mlx5_os.h | 2 +- 3 files changed, 152 insertions(+), 23 deletions(-) diff --git a/drivers/net/mlx5/linux/mlx5_ethdev_os.c b/drivers/net/mlx5/linux/mlx5_ethdev_os.c index 55cae148f0..e3b27818ff 100644 --- a/drivers/net/mlx5/linux/mlx5_ethdev_os.c +++ b/drivers/net/mlx5/linux/mlx5_ethdev_os.c @@ -48,15 +48,20 @@ #include "mlx5_utils.h" /* Get interface index from SubFunction device name. */ -int +unsigned int mlx5_auxiliary_get_ifindex(const char *sf_name) { char if_name[IF_NAMESIZE] = { 0 }; + unsigned int ifindex; if (mlx5_auxiliary_get_child_name(sf_name, "/net", if_name, sizeof(if_name)) != 0) - return -rte_errno; - return if_nametoindex(if_name); + return 0; + + ifindex = if_nametoindex(if_name); + if (ifindex == 0) + rte_errno = errno; + return ifindex; } /** diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c index 83c9774443..ef89510b6a 100644 --- a/drivers/net/mlx5/linux/mlx5_os.c +++ b/drivers/net/mlx5/linux/mlx5_os.c @@ -3151,43 +3151,167 @@ mlx5_os_pci_probe(struct mlx5_common_device *cdev, return ret; } -/* Probe a single SF device on auxiliary bus, no representor support. */ +static int +mlx5_auxiliary_fill_single_spawn_data(struct mlx5_common_device *cdev, + struct rte_device *dev, + struct mlx5_dev_spawn_data *spawn) +{ + unsigned int ifindex; + + /* Probed device is the primary IB device. */ + MLX5_ASSERT(cdev->dev == dev); + /* IB device is single port. */ + MLX5_ASSERT(spawn->max_port == 1); + /* IB device name is provided. */ + MLX5_ASSERT(spawn->phys_dev_name != NULL); + + spawn->phys_port = 1; + ifindex = mlx5_auxiliary_get_ifindex(dev->name); + if (ifindex == 0) { + DRV_LOG(ERR, "dev %s failed to get ethdev ifindex (%s)", + dev->name, rte_strerror(rte_errno)); + return -rte_errno; + } + spawn->ifindex = ifindex; + spawn->cdev = cdev; + spawn->dpdk_dev = dev; + + return 0; +} + +static int +mlx5_auxiliary_fill_switch_spawn_data(struct mlx5_common_device *cdev, + struct rte_device *dev, + int nl_rdma, + int nl_route, + struct mlx5_dev_spawn_data *spawn) +{ + unsigned int dev_ifindex; + int ret; + + /* IB device is multi port. */ + MLX5_ASSERT(spawn->max_port > 1); + /* IB device name is provided. */ + MLX5_ASSERT(spawn->phys_dev_name != NULL); + + dev_ifindex = mlx5_auxiliary_get_ifindex(dev->name); + if (dev_ifindex == 0) { + DRV_LOG(ERR, "dev %s failed to get ethdev ifindex (%s)", + dev->name, rte_strerror(rte_errno)); + return -rte_errno; + } + + for (unsigned int port = 1; port <= spawn->max_port; ++port) { + struct mlx5_switch_info switch_info; + unsigned int ifindex; + + ifindex = mlx5_nl_ifindex(nl_rdma, spawn->phys_dev_name, port, NULL); + if (ifindex == 0) + continue; + + if (ifindex != dev_ifindex) + continue; + + ret = mlx5_nl_switch_info(cdev, nl_route, ifindex, &switch_info); + if (ret < 0) + return ret; + + if (switch_info.name_type != MLX5_PHYS_PORT_NAME_TYPE_VIRTUAL) + continue; + + /* All checks passed. This IB port can be used. */ + spawn->phys_port = port; + spawn->ifindex = ifindex; + spawn->cdev = cdev; + spawn->dpdk_dev = dev; + spawn->info = switch_info; + return 0; + } + + /* If no valid port was found, then probed device cannot be used. */ + rte_errno = ENODEV; + return -rte_errno; +} + +static int +mlx5_auxiliary_fill_spawn_data(struct mlx5_common_device *cdev, + struct rte_device *dev, + struct mlx5_dev_spawn_data *spawn) +{ + const char *phys_dev_name; + unsigned int portnum; + int nl_route = -1; + int nl_rdma = -1; + int ret; + + phys_dev_name = mlx5_os_get_ctx_device_name(cdev->ctx); + if (phys_dev_name == NULL) + return -EINVAL; + + nl_rdma = mlx5_nl_init(NETLINK_RDMA, 0); + if (nl_rdma < 0) { + /* If NETLINK_RDMA is not available, assume single port for fallback. */ + DRV_LOG(DEBUG, + "cdev %s Unable to open NETLINK_RDMA socket, assume single IB port (%s)", + cdev->dev->name, rte_strerror(rte_errno)); + rte_errno = 0; + portnum = 1; + goto fill_spawn_data; + } + + portnum = mlx5_nl_portnum(nl_rdma, phys_dev_name, &cdev->dev_info); + if (portnum == 0) { + ret = -rte_errno; + goto exit; + } + +fill_spawn_data: + memset(spawn, 0, sizeof(*spawn)); + spawn->pf_bond = -1; + spawn->mpesw_port = MLX5_MPESW_PORT_INVALID; + spawn->max_port = portnum; + spawn->phys_dev_name = phys_dev_name; + + if (portnum == 1) { + ret = mlx5_auxiliary_fill_single_spawn_data(cdev, dev, spawn); + } else { + nl_route = mlx5_nl_init(NETLINK_ROUTE, 0); + if (nl_route < 0) { + ret = nl_route; + goto exit; + } + ret = mlx5_auxiliary_fill_switch_spawn_data(cdev, dev, nl_rdma, nl_route, spawn); + } + +exit: + if (nl_route >= 0) + close(nl_route); + if (nl_rdma >= 0) + close(nl_rdma); + return ret; +} + static int mlx5_os_auxiliary_probe(struct mlx5_common_device *cdev, struct rte_device *dev, struct mlx5_kvargs_ctrl *mkvlist) { struct rte_eth_devargs eth_da = { .nb_ports = 0 }; - struct mlx5_dev_spawn_data spawn = { - .pf_bond = -1, - .mpesw_port = MLX5_MPESW_PORT_INVALID, - }; + struct mlx5_dev_spawn_data spawn; struct rte_auxiliary_device *adev = RTE_BUS_DEVICE(dev, *adev); struct rte_eth_dev *eth_dev; int ret = 0; - if (cdev->dev != dev) { - DRV_LOG(ERR, "Invalid device %s used to probe common device %s", - dev->name, cdev->dev->name); - rte_errno = EINVAL; - return -rte_errno; - } /* Parse ethdev devargs. */ ret = mlx5_os_parse_eth_devargs(dev, ð_da); if (ret != 0) return ret; /* Init spawn data. */ - spawn.max_port = 1; - spawn.phys_port = 1; - spawn.phys_dev_name = mlx5_os_get_ctx_device_name(cdev->ctx); - ret = mlx5_auxiliary_get_ifindex(dev->name); + ret = mlx5_auxiliary_fill_spawn_data(cdev, dev, &spawn); if (ret < 0) { - DRV_LOG(ERR, "failed to get ethdev ifindex: %s", dev->name); + DRV_LOG(ERR, "failed to fill spawn data: %s", dev->name); return ret; } - spawn.ifindex = ret; - spawn.cdev = cdev; - spawn.dpdk_dev = dev; /* Spawn device. */ eth_dev = mlx5_dev_spawn(&spawn, ð_da, mkvlist); if (eth_dev == NULL) diff --git a/drivers/net/mlx5/linux/mlx5_os.h b/drivers/net/mlx5/linux/mlx5_os.h index 4d073048fb..c55c9bcefa 100644 --- a/drivers/net/mlx5/linux/mlx5_os.h +++ b/drivers/net/mlx5/linux/mlx5_os.h @@ -13,7 +13,7 @@ #define MLX5_NAMESIZE IF_NAMESIZE -int mlx5_auxiliary_get_ifindex(const char *sf_name); +unsigned int mlx5_auxiliary_get_ifindex(const char *sf_name); enum mlx5_sw_parsing_offloads { -- 2.43.0
